[SCSI] Replace __FUNCTION__ with __func__ in iscsi_tcp.
[linux-2.6/linux-2.6-openrd.git] / mm / memcontrol.c
blobe46451e1d9b793563b08d8ef015927b484b0be1d
1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
20 #include <linux/res_counter.h>
21 #include <linux/memcontrol.h>
22 #include <linux/cgroup.h>
23 #include <linux/mm.h>
24 #include <linux/smp.h>
25 #include <linux/page-flags.h>
26 #include <linux/backing-dev.h>
27 #include <linux/bit_spinlock.h>
28 #include <linux/rcupdate.h>
29 #include <linux/slab.h>
30 #include <linux/swap.h>
31 #include <linux/spinlock.h>
32 #include <linux/fs.h>
33 #include <linux/seq_file.h>
34 #include <linux/vmalloc.h>
36 #include <asm/uaccess.h>
38 struct cgroup_subsys mem_cgroup_subsys;
39 static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
40 static struct kmem_cache *page_cgroup_cache;
43 * Statistics for memory cgroup.
45 enum mem_cgroup_stat_index {
47 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
49 MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */
50 MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */
51 MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */
52 MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */
54 MEM_CGROUP_STAT_NSTATS,
57 struct mem_cgroup_stat_cpu {
58 s64 count[MEM_CGROUP_STAT_NSTATS];
59 } ____cacheline_aligned_in_smp;
61 struct mem_cgroup_stat {
62 struct mem_cgroup_stat_cpu cpustat[NR_CPUS];
66 * For accounting under irq disable, no need for increment preempt count.
68 static void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat *stat,
69 enum mem_cgroup_stat_index idx, int val)
71 int cpu = smp_processor_id();
72 stat->cpustat[cpu].count[idx] += val;
75 static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat,
76 enum mem_cgroup_stat_index idx)
78 int cpu;
79 s64 ret = 0;
80 for_each_possible_cpu(cpu)
81 ret += stat->cpustat[cpu].count[idx];
82 return ret;
86 * per-zone information in memory controller.
89 enum mem_cgroup_zstat_index {
90 MEM_CGROUP_ZSTAT_ACTIVE,
91 MEM_CGROUP_ZSTAT_INACTIVE,
93 NR_MEM_CGROUP_ZSTAT,
96 struct mem_cgroup_per_zone {
98 * spin_lock to protect the per cgroup LRU
100 spinlock_t lru_lock;
101 struct list_head active_list;
102 struct list_head inactive_list;
103 unsigned long count[NR_MEM_CGROUP_ZSTAT];
105 /* Macro for accessing counter */
106 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
108 struct mem_cgroup_per_node {
109 struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES];
112 struct mem_cgroup_lru_info {
113 struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES];
117 * The memory controller data structure. The memory controller controls both
118 * page cache and RSS per cgroup. We would eventually like to provide
119 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
120 * to help the administrator determine what knobs to tune.
122 * TODO: Add a water mark for the memory controller. Reclaim will begin when
123 * we hit the water mark. May be even add a low water mark, such that
124 * no reclaim occurs from a cgroup at it's low water mark, this is
125 * a feature that will be implemented much later in the future.
127 struct mem_cgroup {
128 struct cgroup_subsys_state css;
130 * the counter to account for memory usage
132 struct res_counter res;
134 * Per cgroup active and inactive list, similar to the
135 * per zone LRU lists.
137 struct mem_cgroup_lru_info info;
139 int prev_priority; /* for recording reclaim priority */
141 * statistics.
143 struct mem_cgroup_stat stat;
145 static struct mem_cgroup init_mem_cgroup;
148 * We use the lower bit of the page->page_cgroup pointer as a bit spin
149 * lock. We need to ensure that page->page_cgroup is at least two
150 * byte aligned (based on comments from Nick Piggin). But since
151 * bit_spin_lock doesn't actually set that lock bit in a non-debug
152 * uniprocessor kernel, we should avoid setting it here too.
154 #define PAGE_CGROUP_LOCK_BIT 0x0
155 #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)
156 #define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
157 #else
158 #define PAGE_CGROUP_LOCK 0x0
159 #endif
162 * A page_cgroup page is associated with every page descriptor. The
163 * page_cgroup helps us identify information about the cgroup
165 struct page_cgroup {
166 struct list_head lru; /* per cgroup LRU list */
167 struct page *page;
168 struct mem_cgroup *mem_cgroup;
169 int ref_cnt; /* cached, mapped, migrating */
170 int flags;
172 #define PAGE_CGROUP_FLAG_CACHE (0x1) /* charged as cache */
173 #define PAGE_CGROUP_FLAG_ACTIVE (0x2) /* page is active in this cgroup */
175 static int page_cgroup_nid(struct page_cgroup *pc)
177 return page_to_nid(pc->page);
180 static enum zone_type page_cgroup_zid(struct page_cgroup *pc)
182 return page_zonenum(pc->page);
185 enum charge_type {
186 MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
187 MEM_CGROUP_CHARGE_TYPE_MAPPED,
191 * Always modified under lru lock. Then, not necessary to preempt_disable()
193 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, int flags,
194 bool charge)
196 int val = (charge)? 1 : -1;
197 struct mem_cgroup_stat *stat = &mem->stat;
199 VM_BUG_ON(!irqs_disabled());
200 if (flags & PAGE_CGROUP_FLAG_CACHE)
201 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_CACHE, val);
202 else
203 __mem_cgroup_stat_add_safe(stat, MEM_CGROUP_STAT_RSS, val);
205 if (charge)
206 __mem_cgroup_stat_add_safe(stat,
207 MEM_CGROUP_STAT_PGPGIN_COUNT, 1);
208 else
209 __mem_cgroup_stat_add_safe(stat,
210 MEM_CGROUP_STAT_PGPGOUT_COUNT, 1);
213 static struct mem_cgroup_per_zone *
214 mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid)
216 return &mem->info.nodeinfo[nid]->zoneinfo[zid];
219 static struct mem_cgroup_per_zone *
220 page_cgroup_zoneinfo(struct page_cgroup *pc)
222 struct mem_cgroup *mem = pc->mem_cgroup;
223 int nid = page_cgroup_nid(pc);
224 int zid = page_cgroup_zid(pc);
226 return mem_cgroup_zoneinfo(mem, nid, zid);
229 static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem,
230 enum mem_cgroup_zstat_index idx)
232 int nid, zid;
233 struct mem_cgroup_per_zone *mz;
234 u64 total = 0;
236 for_each_online_node(nid)
237 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
238 mz = mem_cgroup_zoneinfo(mem, nid, zid);
239 total += MEM_CGROUP_ZSTAT(mz, idx);
241 return total;
244 static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
246 return container_of(cgroup_subsys_state(cont,
247 mem_cgroup_subsys_id), struct mem_cgroup,
248 css);
251 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
253 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
254 struct mem_cgroup, css);
257 static inline int page_cgroup_locked(struct page *page)
259 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
262 static void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
264 VM_BUG_ON(!page_cgroup_locked(page));
265 page->page_cgroup = ((unsigned long)pc | PAGE_CGROUP_LOCK);
268 struct page_cgroup *page_get_page_cgroup(struct page *page)
270 return (struct page_cgroup *) (page->page_cgroup & ~PAGE_CGROUP_LOCK);
273 static void lock_page_cgroup(struct page *page)
275 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
278 static int try_lock_page_cgroup(struct page *page)
280 return bit_spin_trylock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
283 static void unlock_page_cgroup(struct page *page)
285 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
288 static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz,
289 struct page_cgroup *pc)
291 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
293 if (from)
294 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
295 else
296 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
298 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, false);
299 list_del_init(&pc->lru);
302 static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz,
303 struct page_cgroup *pc)
305 int to = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
307 if (!to) {
308 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
309 list_add(&pc->lru, &mz->inactive_list);
310 } else {
311 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
312 list_add(&pc->lru, &mz->active_list);
314 mem_cgroup_charge_statistics(pc->mem_cgroup, pc->flags, true);
317 static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
319 int from = pc->flags & PAGE_CGROUP_FLAG_ACTIVE;
320 struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc);
322 if (from)
323 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) -= 1;
324 else
325 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) -= 1;
327 if (active) {
328 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE) += 1;
329 pc->flags |= PAGE_CGROUP_FLAG_ACTIVE;
330 list_move(&pc->lru, &mz->active_list);
331 } else {
332 MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE) += 1;
333 pc->flags &= ~PAGE_CGROUP_FLAG_ACTIVE;
334 list_move(&pc->lru, &mz->inactive_list);
338 int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
340 int ret;
342 task_lock(task);
343 ret = task->mm && mm_match_cgroup(task->mm, mem);
344 task_unlock(task);
345 return ret;
349 * This routine assumes that the appropriate zone's lru lock is already held
351 void mem_cgroup_move_lists(struct page *page, bool active)
353 struct page_cgroup *pc;
354 struct mem_cgroup_per_zone *mz;
355 unsigned long flags;
358 * We cannot lock_page_cgroup while holding zone's lru_lock,
359 * because other holders of lock_page_cgroup can be interrupted
360 * with an attempt to rotate_reclaimable_page. But we cannot
361 * safely get to page_cgroup without it, so just try_lock it:
362 * mem_cgroup_isolate_pages allows for page left on wrong list.
364 if (!try_lock_page_cgroup(page))
365 return;
367 pc = page_get_page_cgroup(page);
368 if (pc) {
369 mz = page_cgroup_zoneinfo(pc);
370 spin_lock_irqsave(&mz->lru_lock, flags);
371 __mem_cgroup_move_lists(pc, active);
372 spin_unlock_irqrestore(&mz->lru_lock, flags);
374 unlock_page_cgroup(page);
378 * Calculate mapped_ratio under memory controller. This will be used in
379 * vmscan.c for deteremining we have to reclaim mapped pages.
381 int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem)
383 long total, rss;
386 * usage is recorded in bytes. But, here, we assume the number of
387 * physical pages can be represented by "long" on any arch.
389 total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L;
390 rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS);
391 return (int)((rss * 100L) / total);
395 * This function is called from vmscan.c. In page reclaiming loop. balance
396 * between active and inactive list is calculated. For memory controller
397 * page reclaiming, we should use using mem_cgroup's imbalance rather than
398 * zone's global lru imbalance.
400 long mem_cgroup_reclaim_imbalance(struct mem_cgroup *mem)
402 unsigned long active, inactive;
403 /* active and inactive are the number of pages. 'long' is ok.*/
404 active = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_ACTIVE);
405 inactive = mem_cgroup_get_all_zonestat(mem, MEM_CGROUP_ZSTAT_INACTIVE);
406 return (long) (active / (inactive + 1));
410 * prev_priority control...this will be used in memory reclaim path.
412 int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem)
414 return mem->prev_priority;
417 void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority)
419 if (priority < mem->prev_priority)
420 mem->prev_priority = priority;
423 void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority)
425 mem->prev_priority = priority;
429 * Calculate # of pages to be scanned in this priority/zone.
430 * See also vmscan.c
432 * priority starts from "DEF_PRIORITY" and decremented in each loop.
433 * (see include/linux/mmzone.h)
436 long mem_cgroup_calc_reclaim_active(struct mem_cgroup *mem,
437 struct zone *zone, int priority)
439 long nr_active;
440 int nid = zone->zone_pgdat->node_id;
441 int zid = zone_idx(zone);
442 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
444 nr_active = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_ACTIVE);
445 return (nr_active >> priority);
448 long mem_cgroup_calc_reclaim_inactive(struct mem_cgroup *mem,
449 struct zone *zone, int priority)
451 long nr_inactive;
452 int nid = zone->zone_pgdat->node_id;
453 int zid = zone_idx(zone);
454 struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid);
456 nr_inactive = MEM_CGROUP_ZSTAT(mz, MEM_CGROUP_ZSTAT_INACTIVE);
457 return (nr_inactive >> priority);
460 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
461 struct list_head *dst,
462 unsigned long *scanned, int order,
463 int mode, struct zone *z,
464 struct mem_cgroup *mem_cont,
465 int active)
467 unsigned long nr_taken = 0;
468 struct page *page;
469 unsigned long scan;
470 LIST_HEAD(pc_list);
471 struct list_head *src;
472 struct page_cgroup *pc, *tmp;
473 int nid = z->zone_pgdat->node_id;
474 int zid = zone_idx(z);
475 struct mem_cgroup_per_zone *mz;
477 BUG_ON(!mem_cont);
478 mz = mem_cgroup_zoneinfo(mem_cont, nid, zid);
479 if (active)
480 src = &mz->active_list;
481 else
482 src = &mz->inactive_list;
485 spin_lock(&mz->lru_lock);
486 scan = 0;
487 list_for_each_entry_safe_reverse(pc, tmp, src, lru) {
488 if (scan >= nr_to_scan)
489 break;
490 page = pc->page;
492 if (unlikely(!PageLRU(page)))
493 continue;
495 if (PageActive(page) && !active) {
496 __mem_cgroup_move_lists(pc, true);
497 continue;
499 if (!PageActive(page) && active) {
500 __mem_cgroup_move_lists(pc, false);
501 continue;
504 scan++;
505 list_move(&pc->lru, &pc_list);
507 if (__isolate_lru_page(page, mode) == 0) {
508 list_move(&page->lru, dst);
509 nr_taken++;
513 list_splice(&pc_list, src);
514 spin_unlock(&mz->lru_lock);
516 *scanned = scan;
517 return nr_taken;
521 * Charge the memory controller for page usage.
522 * Return
523 * 0 if the charge was successful
524 * < 0 if the cgroup is over its limit
526 static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
527 gfp_t gfp_mask, enum charge_type ctype)
529 struct mem_cgroup *mem;
530 struct page_cgroup *pc;
531 unsigned long flags;
532 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
533 struct mem_cgroup_per_zone *mz;
535 if (mem_cgroup_subsys.disabled)
536 return 0;
539 * Should page_cgroup's go to their own slab?
540 * One could optimize the performance of the charging routine
541 * by saving a bit in the page_flags and using it as a lock
542 * to see if the cgroup page already has a page_cgroup associated
543 * with it
545 retry:
546 lock_page_cgroup(page);
547 pc = page_get_page_cgroup(page);
549 * The page_cgroup exists and
550 * the page has already been accounted.
552 if (pc) {
553 VM_BUG_ON(pc->page != page);
554 VM_BUG_ON(pc->ref_cnt <= 0);
556 pc->ref_cnt++;
557 unlock_page_cgroup(page);
558 goto done;
560 unlock_page_cgroup(page);
562 pc = kmem_cache_zalloc(page_cgroup_cache, gfp_mask);
563 if (pc == NULL)
564 goto err;
567 * We always charge the cgroup the mm_struct belongs to.
568 * The mm_struct's mem_cgroup changes on task migration if the
569 * thread group leader migrates. It's possible that mm is not
570 * set, if so charge the init_mm (happens for pagecache usage).
572 if (!mm)
573 mm = &init_mm;
575 rcu_read_lock();
576 mem = mem_cgroup_from_task(rcu_dereference(mm->owner));
578 * For every charge from the cgroup, increment reference count
580 css_get(&mem->css);
581 rcu_read_unlock();
583 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
584 if (!(gfp_mask & __GFP_WAIT))
585 goto out;
587 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
588 continue;
591 * try_to_free_mem_cgroup_pages() might not give us a full
592 * picture of reclaim. Some pages are reclaimed and might be
593 * moved to swap cache or just unmapped from the cgroup.
594 * Check the limit again to see if the reclaim reduced the
595 * current usage of the cgroup before giving up
597 if (res_counter_check_under_limit(&mem->res))
598 continue;
600 if (!nr_retries--) {
601 mem_cgroup_out_of_memory(mem, gfp_mask);
602 goto out;
606 pc->ref_cnt = 1;
607 pc->mem_cgroup = mem;
608 pc->page = page;
609 pc->flags = PAGE_CGROUP_FLAG_ACTIVE;
610 if (ctype == MEM_CGROUP_CHARGE_TYPE_CACHE)
611 pc->flags = PAGE_CGROUP_FLAG_CACHE;
613 lock_page_cgroup(page);
614 if (page_get_page_cgroup(page)) {
615 unlock_page_cgroup(page);
617 * Another charge has been added to this page already.
618 * We take lock_page_cgroup(page) again and read
619 * page->cgroup, increment refcnt.... just retry is OK.
621 res_counter_uncharge(&mem->res, PAGE_SIZE);
622 css_put(&mem->css);
623 kmem_cache_free(page_cgroup_cache, pc);
624 goto retry;
626 page_assign_page_cgroup(page, pc);
628 mz = page_cgroup_zoneinfo(pc);
629 spin_lock_irqsave(&mz->lru_lock, flags);
630 __mem_cgroup_add_list(mz, pc);
631 spin_unlock_irqrestore(&mz->lru_lock, flags);
633 unlock_page_cgroup(page);
634 done:
635 return 0;
636 out:
637 css_put(&mem->css);
638 kmem_cache_free(page_cgroup_cache, pc);
639 err:
640 return -ENOMEM;
643 int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask)
645 return mem_cgroup_charge_common(page, mm, gfp_mask,
646 MEM_CGROUP_CHARGE_TYPE_MAPPED);
649 int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
650 gfp_t gfp_mask)
652 if (!mm)
653 mm = &init_mm;
654 return mem_cgroup_charge_common(page, mm, gfp_mask,
655 MEM_CGROUP_CHARGE_TYPE_CACHE);
659 * Uncharging is always a welcome operation, we never complain, simply
660 * uncharge.
662 void mem_cgroup_uncharge_page(struct page *page)
664 struct page_cgroup *pc;
665 struct mem_cgroup *mem;
666 struct mem_cgroup_per_zone *mz;
667 unsigned long flags;
669 if (mem_cgroup_subsys.disabled)
670 return;
673 * Check if our page_cgroup is valid
675 lock_page_cgroup(page);
676 pc = page_get_page_cgroup(page);
677 if (!pc)
678 goto unlock;
680 VM_BUG_ON(pc->page != page);
681 VM_BUG_ON(pc->ref_cnt <= 0);
683 if (--(pc->ref_cnt) == 0) {
684 mz = page_cgroup_zoneinfo(pc);
685 spin_lock_irqsave(&mz->lru_lock, flags);
686 __mem_cgroup_remove_list(mz, pc);
687 spin_unlock_irqrestore(&mz->lru_lock, flags);
689 page_assign_page_cgroup(page, NULL);
690 unlock_page_cgroup(page);
692 mem = pc->mem_cgroup;
693 res_counter_uncharge(&mem->res, PAGE_SIZE);
694 css_put(&mem->css);
696 kmem_cache_free(page_cgroup_cache, pc);
697 return;
700 unlock:
701 unlock_page_cgroup(page);
705 * Returns non-zero if a page (under migration) has valid page_cgroup member.
706 * Refcnt of page_cgroup is incremented.
708 int mem_cgroup_prepare_migration(struct page *page)
710 struct page_cgroup *pc;
712 if (mem_cgroup_subsys.disabled)
713 return 0;
715 lock_page_cgroup(page);
716 pc = page_get_page_cgroup(page);
717 if (pc)
718 pc->ref_cnt++;
719 unlock_page_cgroup(page);
720 return pc != NULL;
723 void mem_cgroup_end_migration(struct page *page)
725 mem_cgroup_uncharge_page(page);
729 * We know both *page* and *newpage* are now not-on-LRU and PG_locked.
730 * And no race with uncharge() routines because page_cgroup for *page*
731 * has extra one reference by mem_cgroup_prepare_migration.
733 void mem_cgroup_page_migration(struct page *page, struct page *newpage)
735 struct page_cgroup *pc;
736 struct mem_cgroup_per_zone *mz;
737 unsigned long flags;
739 lock_page_cgroup(page);
740 pc = page_get_page_cgroup(page);
741 if (!pc) {
742 unlock_page_cgroup(page);
743 return;
746 mz = page_cgroup_zoneinfo(pc);
747 spin_lock_irqsave(&mz->lru_lock, flags);
748 __mem_cgroup_remove_list(mz, pc);
749 spin_unlock_irqrestore(&mz->lru_lock, flags);
751 page_assign_page_cgroup(page, NULL);
752 unlock_page_cgroup(page);
754 pc->page = newpage;
755 lock_page_cgroup(newpage);
756 page_assign_page_cgroup(newpage, pc);
758 mz = page_cgroup_zoneinfo(pc);
759 spin_lock_irqsave(&mz->lru_lock, flags);
760 __mem_cgroup_add_list(mz, pc);
761 spin_unlock_irqrestore(&mz->lru_lock, flags);
763 unlock_page_cgroup(newpage);
767 * This routine traverse page_cgroup in given list and drop them all.
768 * This routine ignores page_cgroup->ref_cnt.
769 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
771 #define FORCE_UNCHARGE_BATCH (128)
772 static void mem_cgroup_force_empty_list(struct mem_cgroup *mem,
773 struct mem_cgroup_per_zone *mz,
774 int active)
776 struct page_cgroup *pc;
777 struct page *page;
778 int count = FORCE_UNCHARGE_BATCH;
779 unsigned long flags;
780 struct list_head *list;
782 if (active)
783 list = &mz->active_list;
784 else
785 list = &mz->inactive_list;
787 spin_lock_irqsave(&mz->lru_lock, flags);
788 while (!list_empty(list)) {
789 pc = list_entry(list->prev, struct page_cgroup, lru);
790 page = pc->page;
791 get_page(page);
792 spin_unlock_irqrestore(&mz->lru_lock, flags);
793 mem_cgroup_uncharge_page(page);
794 put_page(page);
795 if (--count <= 0) {
796 count = FORCE_UNCHARGE_BATCH;
797 cond_resched();
799 spin_lock_irqsave(&mz->lru_lock, flags);
801 spin_unlock_irqrestore(&mz->lru_lock, flags);
805 * make mem_cgroup's charge to be 0 if there is no task.
806 * This enables deleting this mem_cgroup.
808 static int mem_cgroup_force_empty(struct mem_cgroup *mem)
810 int ret = -EBUSY;
811 int node, zid;
813 if (mem_cgroup_subsys.disabled)
814 return 0;
816 css_get(&mem->css);
818 * page reclaim code (kswapd etc..) will move pages between
819 * active_list <-> inactive_list while we don't take a lock.
820 * So, we have to do loop here until all lists are empty.
822 while (mem->res.usage > 0) {
823 if (atomic_read(&mem->css.cgroup->count) > 0)
824 goto out;
825 for_each_node_state(node, N_POSSIBLE)
826 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
827 struct mem_cgroup_per_zone *mz;
828 mz = mem_cgroup_zoneinfo(mem, node, zid);
829 /* drop all page_cgroup in active_list */
830 mem_cgroup_force_empty_list(mem, mz, 1);
831 /* drop all page_cgroup in inactive_list */
832 mem_cgroup_force_empty_list(mem, mz, 0);
835 ret = 0;
836 out:
837 css_put(&mem->css);
838 return ret;
841 static int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
843 *tmp = memparse(buf, &buf);
844 if (*buf != '\0')
845 return -EINVAL;
848 * Round up the value to the closest page size
850 *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
851 return 0;
854 static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft)
856 return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res,
857 cft->private);
860 static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
861 struct file *file, const char __user *userbuf,
862 size_t nbytes, loff_t *ppos)
864 return res_counter_write(&mem_cgroup_from_cont(cont)->res,
865 cft->private, userbuf, nbytes, ppos,
866 mem_cgroup_write_strategy);
869 static int mem_cgroup_reset(struct cgroup *cont, unsigned int event)
871 struct mem_cgroup *mem;
873 mem = mem_cgroup_from_cont(cont);
874 switch (event) {
875 case RES_MAX_USAGE:
876 res_counter_reset_max(&mem->res);
877 break;
878 case RES_FAILCNT:
879 res_counter_reset_failcnt(&mem->res);
880 break;
882 return 0;
885 static int mem_force_empty_write(struct cgroup *cont, unsigned int event)
887 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont));
890 static const struct mem_cgroup_stat_desc {
891 const char *msg;
892 u64 unit;
893 } mem_cgroup_stat_desc[] = {
894 [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, },
895 [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
896 [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, },
897 [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, },
900 static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
901 struct cgroup_map_cb *cb)
903 struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
904 struct mem_cgroup_stat *stat = &mem_cont->stat;
905 int i;
907 for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) {
908 s64 val;
910 val = mem_cgroup_read_stat(stat, i);
911 val *= mem_cgroup_stat_desc[i].unit;
912 cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
914 /* showing # of active pages */
916 unsigned long active, inactive;
918 inactive = mem_cgroup_get_all_zonestat(mem_cont,
919 MEM_CGROUP_ZSTAT_INACTIVE);
920 active = mem_cgroup_get_all_zonestat(mem_cont,
921 MEM_CGROUP_ZSTAT_ACTIVE);
922 cb->fill(cb, "active", (active) * PAGE_SIZE);
923 cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);
925 return 0;
928 static struct cftype mem_cgroup_files[] = {
930 .name = "usage_in_bytes",
931 .private = RES_USAGE,
932 .read_u64 = mem_cgroup_read,
935 .name = "max_usage_in_bytes",
936 .private = RES_MAX_USAGE,
937 .trigger = mem_cgroup_reset,
938 .read_u64 = mem_cgroup_read,
941 .name = "limit_in_bytes",
942 .private = RES_LIMIT,
943 .write = mem_cgroup_write,
944 .read_u64 = mem_cgroup_read,
947 .name = "failcnt",
948 .private = RES_FAILCNT,
949 .trigger = mem_cgroup_reset,
950 .read_u64 = mem_cgroup_read,
953 .name = "force_empty",
954 .trigger = mem_force_empty_write,
957 .name = "stat",
958 .read_map = mem_control_stat_show,
962 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
964 struct mem_cgroup_per_node *pn;
965 struct mem_cgroup_per_zone *mz;
966 int zone, tmp = node;
968 * This routine is called against possible nodes.
969 * But it's BUG to call kmalloc() against offline node.
971 * TODO: this routine can waste much memory for nodes which will
972 * never be onlined. It's better to use memory hotplug callback
973 * function.
975 if (!node_state(node, N_NORMAL_MEMORY))
976 tmp = -1;
977 pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp);
978 if (!pn)
979 return 1;
981 mem->info.nodeinfo[node] = pn;
982 memset(pn, 0, sizeof(*pn));
984 for (zone = 0; zone < MAX_NR_ZONES; zone++) {
985 mz = &pn->zoneinfo[zone];
986 INIT_LIST_HEAD(&mz->active_list);
987 INIT_LIST_HEAD(&mz->inactive_list);
988 spin_lock_init(&mz->lru_lock);
990 return 0;
993 static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node)
995 kfree(mem->info.nodeinfo[node]);
998 static struct mem_cgroup *mem_cgroup_alloc(void)
1000 struct mem_cgroup *mem;
1002 if (sizeof(*mem) < PAGE_SIZE)
1003 mem = kmalloc(sizeof(*mem), GFP_KERNEL);
1004 else
1005 mem = vmalloc(sizeof(*mem));
1007 if (mem)
1008 memset(mem, 0, sizeof(*mem));
1009 return mem;
1012 static void mem_cgroup_free(struct mem_cgroup *mem)
1014 if (sizeof(*mem) < PAGE_SIZE)
1015 kfree(mem);
1016 else
1017 vfree(mem);
1021 static struct cgroup_subsys_state *
1022 mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
1024 struct mem_cgroup *mem;
1025 int node;
1027 if (unlikely((cont->parent) == NULL)) {
1028 mem = &init_mem_cgroup;
1029 page_cgroup_cache = KMEM_CACHE(page_cgroup, SLAB_PANIC);
1030 } else {
1031 mem = mem_cgroup_alloc();
1032 if (!mem)
1033 return ERR_PTR(-ENOMEM);
1036 res_counter_init(&mem->res);
1038 for_each_node_state(node, N_POSSIBLE)
1039 if (alloc_mem_cgroup_per_zone_info(mem, node))
1040 goto free_out;
1042 return &mem->css;
1043 free_out:
1044 for_each_node_state(node, N_POSSIBLE)
1045 free_mem_cgroup_per_zone_info(mem, node);
1046 if (cont->parent != NULL)
1047 mem_cgroup_free(mem);
1048 return ERR_PTR(-ENOMEM);
1051 static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss,
1052 struct cgroup *cont)
1054 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1055 mem_cgroup_force_empty(mem);
1058 static void mem_cgroup_destroy(struct cgroup_subsys *ss,
1059 struct cgroup *cont)
1061 int node;
1062 struct mem_cgroup *mem = mem_cgroup_from_cont(cont);
1064 for_each_node_state(node, N_POSSIBLE)
1065 free_mem_cgroup_per_zone_info(mem, node);
1067 mem_cgroup_free(mem_cgroup_from_cont(cont));
1070 static int mem_cgroup_populate(struct cgroup_subsys *ss,
1071 struct cgroup *cont)
1073 if (mem_cgroup_subsys.disabled)
1074 return 0;
1075 return cgroup_add_files(cont, ss, mem_cgroup_files,
1076 ARRAY_SIZE(mem_cgroup_files));
1079 static void mem_cgroup_move_task(struct cgroup_subsys *ss,
1080 struct cgroup *cont,
1081 struct cgroup *old_cont,
1082 struct task_struct *p)
1084 struct mm_struct *mm;
1085 struct mem_cgroup *mem, *old_mem;
1087 if (mem_cgroup_subsys.disabled)
1088 return;
1090 mm = get_task_mm(p);
1091 if (mm == NULL)
1092 return;
1094 mem = mem_cgroup_from_cont(cont);
1095 old_mem = mem_cgroup_from_cont(old_cont);
1097 if (mem == old_mem)
1098 goto out;
1101 * Only thread group leaders are allowed to migrate, the mm_struct is
1102 * in effect owned by the leader
1104 if (!thread_group_leader(p))
1105 goto out;
1107 out:
1108 mmput(mm);
1111 struct cgroup_subsys mem_cgroup_subsys = {
1112 .name = "memory",
1113 .subsys_id = mem_cgroup_subsys_id,
1114 .create = mem_cgroup_create,
1115 .pre_destroy = mem_cgroup_pre_destroy,
1116 .destroy = mem_cgroup_destroy,
1117 .populate = mem_cgroup_populate,
1118 .attach = mem_cgroup_move_task,
1119 .early_init = 0,